An AI visibility audit answers four questions about any brand — yours, a client's, a prospect's: Does AI name them when buyers ask? Against whom? Based on which sources? And what, specifically, should they do about it?
This is the complete manual process — the same structure our product automates. Budget about 90 minutes for a first pass. Nothing here needs a tool; everything needs honesty.
Set up a spreadsheet with five tabs (this structure is the free template — copy it as written):
Tab 1 · Questions — 10–15 buying-intent prompts: best-of, vs, alternatives, who-should-I-hire (selection rules).
Tab 2 · Mentions grid — questions × brands; one sheet per engine you test.
Tab 3 · Sources — every URL the engines cite, with a count of how often each domain recurs.
Tab 4 · Framing — the actual sentences used about the audited brand and top competitors.
Tab 5 · Actions — the output: gaps ranked by impact × effort.
The five-tab audit template
Each tab feeds the next — the last one is the deliverable
Tab 5 is the whole point — an audit that ends in a score instead of a to-do list wasn't an audit, it was a screenshot.
Step 1 — ask everything, twice. Run all questions through at least ChatGPT, Gemini, and Perplexity — the engines genuinely disagree1 — and ideally on two different days, because single runs are unreliable.2 Tick every brand named in the grid.
Step 2 — score the headline numbers.Visibility rate (% of answers naming the brand), share of voice (brand's mentions ÷ all mentions), and the same for each competitor. These are the before-photo.
Step 3 — harvest sources.Collect citations from every answer (Perplexity prints them; elsewhere ask “which sources support this? cite them”). The domains recurring across questions are the category's kingmakers — expect a short list of listicles, review platforms, and threads.3
Step 4 — capture the framing.Copy the exact sentences about the audited brand. Recurring adjectives are the corpus's verdict (what feeds them and how to move them).
For every question where competitors show and the brand doesn't, trace it to the sources tab: which cited pages mention the winners but not you? Each one becomes an action — pitch this listicle, build reviews here, answer that thread, publish a direct answer page for this question (structured to be quotable: statistics, sourced claims, clean headings — the patterns measured to lift AI visibility4). Score each 1–5 on impact and effort; the high-impact/low-effort corner is the first sprint. A good audit ends with ten concrete URLs and owners, not a score.
An audit is a photograph of a moving subject. The answers will be different next month — sometimes next week — which is exactly why one-off audit scores mislead.2The audit's real job is to start the loop: baseline → act → re-measure → attribute. Re-run the same template monthly (an hour once the sheet exists), and the trend becomes the scoreboard that makes every action accountable.
And if 90 minutes a month is 90 minutes you don't have: this template is literally what Beacon automates — the questions, the grid across five engines, the source harvest, the sentiment classification, and the ranked actions, refreshed daily. The free check runs the first pass on any brand in about a minute. Either way — spreadsheet or software — run the audit. Every brand has a number; the ones who know theirs get to change it.
The free check runs this exact audit on any brand — five engines, sources, and gaps — while your coffee is still hot.